Skip to content

Generating Scrutiny: FTC Outlines Competition Concerns in Generative AI

Help! The Government Seized My Privileged Documents! What a Recent Fifth Circuit Case Says About How to Respond Background Image

On June 29, 2023, the Federal Trade Commission’s (“FTC”) Bureau of Competition and Office of Technology published a joint blog post that provided deeper insight into the FTC’s focus on competition in generative artificial intelligence (“AI”). Titled “Generative AI Raises Competition Concerns,” the post outlines the FTC’s views regarding the “essential technical building blocks” of generative AI and potential competition concerns raised by the technology. This is the FTC’s clearest signal of its intent to scrutinize the emerging AI sector. Of course, the concerns expressed in a blog post do not necessarily suggest that the FTC could successfully litigate a case in court, and the FTC has not in most cases explained how the practices about which it has expressed concern would actually violate the antitrust laws. Nonetheless, a pre-litigation FTC investigation, which does not require the FTC to satisfy any particular antitrust standard, can be burdensome and costly for companies.

According to the FTC, there are three essential building blocks of generative AI that give incumbent firms an advantage over emerging competitors:

  • Data. The FTC argues that “collecting a large and diverse corpus of data” can be harder for new entrants because incumbents have accumulated large amounts of user data over years and have “developed and honed proprietary data collection tools.” According to the FTC, long term data collected by “digital platforms” and in “specialized” or “highly regulated” domains (the FTC cited health and financial data as examples) are particularly difficult for new players to collect.
  • Talent. The FTC argues that labor expertise and deep mastery of AI skill sets required for cutting-edge generative AI development is key to development and market competition. The FTC cited its proposed rulemaking banning all non-competes as relevant to ensuring that specialized AI talent is not hindered in mobility. The FTC’s proposed rule banning non-competes is, of course, just that — a proposal — and under existing law there remain many situations where non-competes are justified.
  • Computational Resources. The FTC argues that large scale generative AI requires significant computational infrastructure (known as “compute”) and that access to this compute can be expensive and concentrated among a small number of companies. The FTC cited its challenge of and the subsequent abandonment of Nvidia’s proposed acquisition of ARM as evidence of this concern. In addition, the FTC discussed the effects of competition in cloud computing on generative AI development in its recent workshop and request for information on cloud computing business practices.

Of course, merely having an advantage as an incumbent is not illegal, and the FTC has not articulated how it believes a data, talent, or computational advantage would amount to anticompetitive foreclosure. The rapid pace of investment and new entry among generative AI firms suggests that actual market participants do not believe that entry is futile because of incumbents’ advantages in these areas.

In its blog post, the FTC stated a preference for open-source development of generative AI and derided proprietary ecosystems as limiting access to the building blocks of generative AI. However, development of generative AI, the FTC warned of the risks of “open first, closed later” tactics that can “lock-in customers and lock-out competition.”

The FTC outlined multiple “possible unfair methods of competition” that it viewed as particularly relevant in the context of generative AI:

  • Bundling and tying. The FTC argues that incumbents may link together new generative AI apps with existing products to harm competition. Of course, bundling can create well-recognized technical and cost-based efficiencies, which the FTC appears to ignore for purposes of this blog post.
  • Cloud exclusivity arrangements. The FTC argues that incumbents who offer computational infrastructure and cloud services can harm competition by requiring exclusivity or preferential treatment on the part of a generative AI startup. Again, the FTC does not explain how foreclosure would occur when there is strong competition both among cloud infrastructure providers and among generative AI companies, nor does it acknowledge well-recognized justifications for exclusivity.
  • Mergers and acquisitions. The FTC argues that incumbents may harm competition by acquiring competing generative AI developers or critical inputs to generative AI development.

The FTC’s blog makes clear that it seeks to apply “vigorous law enforcement” in the context of the generative AI ecosystem. The FTC has already made good on this promise in the consumer protection context as the FTC recently opened an expansive investigation into OpenAI’s generative AI models with a focus on privacy, data security, and use of consumer information in model training. In a time where competition regulators around the world are focusing attention on generative AI, including the European Commission and the UK Competition and Markets Authority, the FTC’s blog post emphasizes the importance of accounting for the competition risks that may be implicated by generative AI.

This information is provided by Vinson & Elkins LLP for educational and informational purposes only and is not intended, nor should it be construed, as legal advice.